Implementing a Hybrid Deep Learning Approach to Achieve Classic Handwritten Alphanumeric MODI Recognition

Abstract

MODI, synonymous with the Devanagari script, is an ancient script from the 17th century used by the Maratha empire as a symbol of culture and power to propagate Marathi. Due to a decline in its usage, absence of quality script database and an unavailability of good literature, identification and translation of MODI script is demanding. The present work deals with a novel study on the recognition of MODI characters and numerals by using Convolutional Neural Network (CNN) architecture. By using a traditional machine learning classifier, classification is performed, and then through a comparative analysis of Random Forest and XGBoost, the study achieves recognition accuracy of 92% for characters and 93.3% for numerals.